The OnlyFans AI chatbot rule, 2022 to 2026

For roughly two years, one sentence has done all the work in this argument: "You cannot use an AI chatbot to write chats or direct messages." It has been quoted in newsrooms, in vendor marketing, in agency contracts and in forum arguments, and almost always with the same attribution: the OnlyFans terms of service.

That attribution is wrong, and the correction is more interesting than the error.

What Reuters published

On 30 July 2024, Reuters published "AI bots talk dirty so OnlyFans stars don't have to", by Andrew R.C. Marshall, Jason Szep and Linda So. The relevant passage reads:

OnlyFans' terms of service explicitly prohibit the use of AI chatbots by its creators. "You cannot use an AI chatbot to write chats or direct messages," the terms say.

The investigation was about software vendors selling AI chat for creator accounts. Reuters also recorded the argument the vendors were already making about the rule: one founder said his software stayed on the right side of it because a human still had to press send, while a competitor said its product could compose and send without human help. Reuters wrote that OnlyFans declined to answer its questions about the use of AI on the site, and that the company has said its terms make creators legally responsible for any transaction with a subscriber.

What the terms of service actually contain

The desk read the live Terms of Use on 2026-09-14. The page carries the footer "Last updated: August 2024". The strings "chatbot", "chat bot" and "automated" appear zero times in it. The Standard Contract between Fan and Creator, also last updated August 2024, contains zero occurrences of "chatbot" or "chat bot".

Artificial intelligence appears in the Terms twice, and neither instance is about chat:

AI Generated Content: AI Generated content must comply with our Terms of Service and must be clearly and conspicuously captioned as AI Generated Content with a signifier such as #ai, or #AIGenerated.

and, describing how OnlyFans reviews content,

we reserve the right to conduct these reviews with technology tools such as classifiers, and Artificial Intelligence and Machine Learning-enabled tools to detect and remove Content that violates our policies.

The archive sweep behind this entry was done by OnlyChatAI and published on 2026-09-13. It reports checking 297 archived captures of the terms and the fan and creator contract, covering February 2022 to August 2026, of which 292 were readable, and finding the quoted sentence in none of them at any date. It also records that the terms were revised exactly once in that period, from a December 2021 text that stood until 25 August 2024 to the August 2024 text that went live on 4 September 2024, five weeks after the Reuters piece.

That publisher sells AI chatting software and says so on the page. The desk did not re-run the archive sweep. It did verify the part that can be checked live, which is that the sentence is not in the current terms, and the part of the archive claim that is cheap to test, which is the state of the archived help article.

Why the Reuters phrasing was defensible anyway

The terms define the agreement broadly. The live page reads:

the legally binding agreement between you and us which consists of: (i) these Terms of Use; (ii) Acceptable Use Policy; (iii) Referral Program Terms; (iv) Platform to Business Regulation Terms; (v) Complaints Policy; (vi) Appeals Policy; (vii) UK VAT Policy; and (viii) Community Guidelines

The Community Guidelines are part of the contract. A rule published there is a contractual rule, and "the terms say" is a fair shorthand for it. What is not fair is what happened next: the sentence travelled without its address, and everyone who repeated it sent readers to a document that never contained it.

Where the rule lived

In the Community Guidelines, in the Sharing Content section, at a help centre article numbered 196/205/213. Its title was "Can I use AI chatbots or AI-generated chat content in chats or direct messages?" and the answer was one word followed by one sentence:

No. You cannot use an AI chatbot to write chats or direct messages on OnlyFans.

The desk cannot open that page. On 2026-09-14 the live URL returns a 17,669 byte application shell with no article text in it, identically for a browser, for Googlebot and for Bingbot, and the content is loaded from an endpoint that requires request signing. The Internet Archive holds exactly one capture of that URL, from 5 April 2025, and it is a 403.

What can be corroborated is the wording, through a third party mirror that reproduces the article, cites the Community Guidelines as its source and links the same help URL. That page is self dated 25 November 2025 and its Russian language version has archive captures from 21 January 2026 and 12 April 2026, both carrying the same answer. A mirror is weaker evidence than the original. It is what exists.

What the help centre says now

The current text of the article reads:

OnlyFans does not offer an internal AI chatbot tool for Creators. OnlyFans does not have a public facing API and Creators may not directly integrate AI chat bots into the OnlyFans platform.

That wording comes from a dated print of the page supplied by the investigation's authors, who sell AI chatting software and disclose that interest. The print carries a timestamp of 13 September 2026 and covers help articles 196, 205 and 213, and the desk has read it. What the desk has not done, and cannot do, is open the live page itself, for the same reason it could not read the old version: the help centre does not serve its text to anything that is not the OnlyFans application. No archive copy and no search engine cache carries those two sentences. So this entry reports a dated print, not a page it retrieved, and says which it is.

If the wording is accurate, the change is substantial. The old sentence was a prohibition on a practice. The new one is two statements of fact about the product: there is no built in tool, there is no public API, and integration is not available. Whether a creator may have messages written for them by software running outside the platform is not addressed by either sentence.

The chronology

What this entry does not say

It does not say that any tool is allowed, compliant or safe on OnlyFans. That is not a question a desk can answer from public documents, and the only party who can answer it has declined to. What the desk can say is narrower and firmer: the sentence that has been quoted as a term of service was not in the terms of service, it was in a help article, that help article is now unreadable from outside the platform, and the last wording anyone has printed from it is not a prohibition but a statement about an API that does not exist.

The Acceptable Use Policy is the one remaining gap. Its page returns the same empty shell, and the archive captures since August 2024 are reported to be Cloudflare interstitials, so the desk has not read its current text and makes no claim about it.